blob: d8838dedb7a4cd06f54e2fa8a522b6c4bb15091e [file] [log] [blame]
Rusty Russell48925e32009-09-24 09:59:20 -06001/* A network driver using virtio.
Rusty Russell296f96f2007-10-22 11:03:37 +10002 *
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
Jeff Kirsheradf8d3f2013-12-06 06:28:47 -080016 * along with this program; if not, see <http://www.gnu.org/licenses/>.
Rusty Russell296f96f2007-10-22 11:03:37 +100017 */
18//#define DEBUG
19#include <linux/netdevice.h>
20#include <linux/etherdevice.h>
Herbert Xua9ea3fc2008-04-18 11:21:42 +080021#include <linux/ethtool.h>
Rusty Russell296f96f2007-10-22 11:03:37 +100022#include <linux/module.h>
23#include <linux/virtio.h>
24#include <linux/virtio_net.h>
25#include <linux/scatterlist.h>
Alex Williamsone918085a2009-01-25 18:06:26 -080026#include <linux/if_vlan.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090027#include <linux/slab.h>
Wanlong Gao8de4b2f2013-01-24 23:51:31 +000028#include <linux/cpu.h>
Michael Daltonab7db912014-01-16 22:23:27 -080029#include <linux/average.h>
Jason Wang91815632014-07-23 16:33:55 +080030#include <net/busy_poll.h>
Rusty Russell296f96f2007-10-22 11:03:37 +100031
Amerigo Wangd34710e2013-05-09 19:50:51 +000032static int napi_weight = NAPI_POLL_WEIGHT;
Dor Laor6c0cd7c2007-12-16 15:19:43 +020033module_param(napi_weight, int, 0444);
34
Rusty Russelleb939922011-12-19 14:08:01 +000035static bool csum = true, gso = true;
Rusty Russell34a48572008-02-04 23:50:02 -050036module_param(csum, bool, 0444);
37module_param(gso, bool, 0444);
38
Rusty Russell296f96f2007-10-22 11:03:37 +100039/* FIXME: MTU in config. */
Michael Dalton5061de32013-11-14 10:41:04 -080040#define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -080041#define GOOD_COPY_LEN 128
Rusty Russell296f96f2007-10-22 11:03:37 +100042
Johannes Berg5377d7582015-08-19 09:48:40 +020043/* RX packet size EWMA. The average packet size is used to determine the packet
44 * buffer size when refilling RX rings. As the entire RX ring may be refilled
45 * at once, the weight is chosen so that the EWMA will be insensitive to short-
46 * term, transient changes in packet size.
Michael Daltonab7db912014-01-16 22:23:27 -080047 */
Johannes Berg5377d7582015-08-19 09:48:40 +020048DECLARE_EWMA(pkt_len, 1, 64)
Michael Daltonab7db912014-01-16 22:23:27 -080049
50/* Minimum alignment for mergeable packet buffers. */
51#define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
52
Rick Jones66846042011-11-14 14:17:08 +000053#define VIRTNET_DRIVER_VERSION "1.0.0"
Alex Williamson2a41f712009-02-04 09:02:34 +000054
stephen hemminger3fa2a1d2011-06-15 06:36:29 +000055struct virtnet_stats {
Eric Dumazet83a27052012-06-05 22:35:24 +000056 struct u64_stats_sync tx_syncp;
57 struct u64_stats_sync rx_syncp;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +000058 u64 tx_bytes;
59 u64 tx_packets;
60
61 u64 rx_bytes;
62 u64 rx_packets;
63};
64
Jason Wange9d74172012-12-07 07:04:55 +000065/* Internal representation of a send virtqueue */
66struct send_queue {
67 /* Virtqueue associated with this send _queue */
68 struct virtqueue *vq;
69
70 /* TX: fragments + linear part + virtio header */
71 struct scatterlist sg[MAX_SKB_FRAGS + 2];
Jason Wang986a4f42012-12-07 07:04:56 +000072
73 /* Name of the send queue: output.$index */
74 char name[40];
Jason Wange9d74172012-12-07 07:04:55 +000075};
76
77/* Internal representation of a receive virtqueue */
78struct receive_queue {
79 /* Virtqueue associated with this receive_queue */
80 struct virtqueue *vq;
81
Rusty Russell296f96f2007-10-22 11:03:37 +100082 struct napi_struct napi;
83
Jason Wange9d74172012-12-07 07:04:55 +000084 /* Chain pages by the private ptr. */
85 struct page *pages;
86
Michael Daltonab7db912014-01-16 22:23:27 -080087 /* Average packet length for mergeable receive buffers. */
Johannes Berg5377d7582015-08-19 09:48:40 +020088 struct ewma_pkt_len mrg_avg_pkt_len;
Michael Daltonab7db912014-01-16 22:23:27 -080089
Michael Daltonfb518792014-01-16 22:23:26 -080090 /* Page frag for packet buffer allocation. */
91 struct page_frag alloc_frag;
92
Jason Wange9d74172012-12-07 07:04:55 +000093 /* RX: fragments + linear part + virtio header */
94 struct scatterlist sg[MAX_SKB_FRAGS + 2];
Jason Wang986a4f42012-12-07 07:04:56 +000095
96 /* Name of this receive queue: input.$index */
97 char name[40];
Jason Wange9d74172012-12-07 07:04:55 +000098};
99
100struct virtnet_info {
101 struct virtio_device *vdev;
102 struct virtqueue *cvq;
103 struct net_device *dev;
Jason Wang986a4f42012-12-07 07:04:56 +0000104 struct send_queue *sq;
105 struct receive_queue *rq;
Jason Wange9d74172012-12-07 07:04:55 +0000106 unsigned int status;
107
Jason Wang986a4f42012-12-07 07:04:56 +0000108 /* Max # of queue pairs supported by the device */
109 u16 max_queue_pairs;
110
111 /* # of queue pairs currently used by the driver */
112 u16 curr_queue_pairs;
113
Herbert Xu97402b92008-04-18 11:24:27 +0800114 /* I like... big packets and I cannot lie! */
115 bool big_packets;
116
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800117 /* Host will merge rx buffers for big packets (shake it! shake it!) */
118 bool mergeable_rx_bufs;
119
Jason Wang986a4f42012-12-07 07:04:56 +0000120 /* Has control virtqueue */
121 bool has_cvq;
122
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930123 /* Host can handle any s/g split between our header and packet data */
124 bool any_header_sg;
125
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300126 /* Packet virtio header size */
127 u8 hdr_len;
128
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000129 /* Active statistics */
130 struct virtnet_stats __percpu *stats;
131
Rusty Russell3161e452009-08-26 12:22:32 -0700132 /* Work struct for refilling if we run low on memory. */
133 struct delayed_work refill;
134
Jason Wang586d17c2012-04-11 20:43:52 +0000135 /* Work struct for config space updates */
136 struct work_struct config_work;
137
Jason Wang986a4f42012-12-07 07:04:56 +0000138 /* Does the affinity hint is set for virtqueues? */
139 bool affinity_hint_set;
Wanlong Gao47be2472013-01-24 23:51:29 +0000140
Wanlong Gao8de4b2f2013-01-24 23:51:31 +0000141 /* CPU hot plug notifier */
142 struct notifier_block nb;
Rusty Russell296f96f2007-10-22 11:03:37 +1000143};
144
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000145struct padded_vnet_hdr {
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300146 struct virtio_net_hdr_mrg_rxbuf hdr;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000147 /*
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300148 * hdr is in a separate sg buffer, and data sg buffer shares same page
149 * with this header sg. This padding makes next sg 16 byte aligned
150 * after the header.
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000151 */
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300152 char padding[4];
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000153};
154
Jason Wang986a4f42012-12-07 07:04:56 +0000155/* Converting between virtqueue no. and kernel tx/rx queue no.
156 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
157 */
158static int vq2txq(struct virtqueue *vq)
159{
Rusty Russell9d0ca6e2013-03-21 14:17:34 +0000160 return (vq->index - 1) / 2;
Jason Wang986a4f42012-12-07 07:04:56 +0000161}
162
163static int txq2vq(int txq)
164{
165 return txq * 2 + 1;
166}
167
168static int vq2rxq(struct virtqueue *vq)
169{
Rusty Russell9d0ca6e2013-03-21 14:17:34 +0000170 return vq->index / 2;
Jason Wang986a4f42012-12-07 07:04:56 +0000171}
172
173static int rxq2vq(int rxq)
174{
175 return rxq * 2;
176}
177
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300178static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +1000179{
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300180 return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
Rusty Russell296f96f2007-10-22 11:03:37 +1000181}
182
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000183/*
184 * private is used to chain pages for big packets, put the whole
185 * most recent used list in the beginning for reuse
186 */
Jason Wange9d74172012-12-07 07:04:55 +0000187static void give_pages(struct receive_queue *rq, struct page *page)
Rusty Russellfb6813f2008-07-25 12:06:01 -0500188{
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000189 struct page *end;
190
Jason Wange9d74172012-12-07 07:04:55 +0000191 /* Find end of list, sew whole thing into vi->rq.pages. */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000192 for (end = page; end->private; end = (struct page *)end->private);
Jason Wange9d74172012-12-07 07:04:55 +0000193 end->private = (unsigned long)rq->pages;
194 rq->pages = page;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500195}
196
Jason Wange9d74172012-12-07 07:04:55 +0000197static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
Rusty Russellfb6813f2008-07-25 12:06:01 -0500198{
Jason Wange9d74172012-12-07 07:04:55 +0000199 struct page *p = rq->pages;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500200
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000201 if (p) {
Jason Wange9d74172012-12-07 07:04:55 +0000202 rq->pages = (struct page *)p->private;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000203 /* clear private here, it is used to chain pages */
204 p->private = 0;
205 } else
Rusty Russellfb6813f2008-07-25 12:06:01 -0500206 p = alloc_page(gfp_mask);
207 return p;
208}
209
Jason Wange9d74172012-12-07 07:04:55 +0000210static void skb_xmit_done(struct virtqueue *vq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000211{
Jason Wange9d74172012-12-07 07:04:55 +0000212 struct virtnet_info *vi = vq->vdev->priv;
Rusty Russell296f96f2007-10-22 11:03:37 +1000213
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500214 /* Suppress further interrupts. */
Jason Wange9d74172012-12-07 07:04:55 +0000215 virtqueue_disable_cb(vq);
Rusty Russell11a3a152008-05-26 17:48:13 +1000216
Rusty Russell363f1512008-06-08 20:51:55 +1000217 /* We were probably waiting for more output buffers. */
Jason Wang986a4f42012-12-07 07:04:56 +0000218 netif_wake_subqueue(vi->dev, vq2txq(vq));
Rusty Russell296f96f2007-10-22 11:03:37 +1000219}
220
Michael Daltonab7db912014-01-16 22:23:27 -0800221static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
222{
223 unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
224 return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
225}
226
227static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
228{
229 return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
230
231}
232
233static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
234{
235 unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
236 return (unsigned long)buf | (size - 1);
237}
238
Mike Waychison34646452012-01-04 12:52:32 +0000239/* Called from bottom half context */
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300240static struct sk_buff *page_to_skb(struct virtnet_info *vi,
241 struct receive_queue *rq,
Michael Dalton2613af02013-10-28 15:44:18 -0700242 struct page *page, unsigned int offset,
243 unsigned int len, unsigned int truesize)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000244{
245 struct sk_buff *skb;
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300246 struct virtio_net_hdr_mrg_rxbuf *hdr;
Michael Dalton2613af02013-10-28 15:44:18 -0700247 unsigned int copy, hdr_len, hdr_padded_len;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000248 char *p;
249
Michael Dalton2613af02013-10-28 15:44:18 -0700250 p = page_address(page) + offset;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000251
252 /* copy small packet so we can reuse these pages for small data */
253 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
254 if (unlikely(!skb))
255 return NULL;
256
257 hdr = skb_vnet_hdr(skb);
258
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300259 hdr_len = vi->hdr_len;
260 if (vi->mergeable_rx_bufs)
261 hdr_padded_len = sizeof *hdr;
262 else
Michael Dalton2613af02013-10-28 15:44:18 -0700263 hdr_padded_len = sizeof(struct padded_vnet_hdr);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000264
265 memcpy(hdr, p, hdr_len);
266
267 len -= hdr_len;
Michael Dalton2613af02013-10-28 15:44:18 -0700268 offset += hdr_padded_len;
269 p += hdr_padded_len;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000270
271 copy = len;
272 if (copy > skb_tailroom(skb))
273 copy = skb_tailroom(skb);
274 memcpy(skb_put(skb, copy), p, copy);
275
276 len -= copy;
277 offset += copy;
278
Michael Dalton2613af02013-10-28 15:44:18 -0700279 if (vi->mergeable_rx_bufs) {
280 if (len)
281 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
282 else
283 put_page(page);
284 return skb;
285 }
286
Sasha Levine878d782011-09-28 04:40:54 +0000287 /*
288 * Verify that we can indeed put this data into a skb.
289 * This is here to handle cases when the device erroneously
290 * tries to receive more than is possible. This is usually
291 * the case of a broken device.
292 */
293 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
Amerigo Wangbe443892012-11-08 17:47:28 +0000294 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
Sasha Levine878d782011-09-28 04:40:54 +0000295 dev_kfree_skb(skb);
296 return NULL;
297 }
Michael Dalton2613af02013-10-28 15:44:18 -0700298 BUG_ON(offset >= PAGE_SIZE);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000299 while (len) {
Michael Dalton2613af02013-10-28 15:44:18 -0700300 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
301 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
302 frag_size, truesize);
303 len -= frag_size;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000304 page = (struct page *)page->private;
305 offset = 0;
306 }
307
308 if (page)
Jason Wange9d74172012-12-07 07:04:55 +0000309 give_pages(rq, page);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000310
311 return skb;
312}
313
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300314static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200315{
316 struct sk_buff * skb = buf;
317
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300318 len -= vi->hdr_len;
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200319 skb_trim(skb, len);
320
321 return skb;
322}
323
324static struct sk_buff *receive_big(struct net_device *dev,
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300325 struct virtnet_info *vi,
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200326 struct receive_queue *rq,
327 void *buf,
328 unsigned int len)
329{
330 struct page *page = buf;
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300331 struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200332
333 if (unlikely(!skb))
334 goto err;
335
336 return skb;
337
338err:
339 dev->stats.rx_dropped++;
340 give_pages(rq, page);
341 return NULL;
342}
343
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200344static struct sk_buff *receive_mergeable(struct net_device *dev,
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +0200345 struct virtnet_info *vi,
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200346 struct receive_queue *rq,
Michael Daltonab7db912014-01-16 22:23:27 -0800347 unsigned long ctx,
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200348 unsigned int len)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000349{
Michael Daltonab7db912014-01-16 22:23:27 -0800350 void *buf = mergeable_ctx_to_buf_address(ctx);
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300351 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
352 u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200353 struct page *page = virt_to_head_page(buf);
354 int offset = buf - page_address(page);
Michael Daltonab7db912014-01-16 22:23:27 -0800355 unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
356
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300357 struct sk_buff *head_skb = page_to_skb(vi, rq, page, offset, len,
358 truesize);
Michael Dalton2613af02013-10-28 15:44:18 -0700359 struct sk_buff *curr_skb = head_skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000360
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200361 if (unlikely(!curr_skb))
362 goto err_skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000363 while (--num_buf) {
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200364 int num_skb_frags;
365
Michael Daltonab7db912014-01-16 22:23:27 -0800366 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
367 if (unlikely(!ctx)) {
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200368 pr_debug("%s: rx error: %d buffers out of %d missing\n",
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +0200369 dev->name, num_buf,
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300370 virtio16_to_cpu(vi->vdev,
371 hdr->num_buffers));
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200372 dev->stats.rx_length_errors++;
373 goto err_buf;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000374 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200375
Michael Daltonab7db912014-01-16 22:23:27 -0800376 buf = mergeable_ctx_to_buf_address(ctx);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200377 page = virt_to_head_page(buf);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200378
379 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
Michael Dalton2613af02013-10-28 15:44:18 -0700380 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
381 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200382
383 if (unlikely(!nskb))
384 goto err_skb;
Michael Dalton2613af02013-10-28 15:44:18 -0700385 if (curr_skb == head_skb)
386 skb_shinfo(curr_skb)->frag_list = nskb;
387 else
388 curr_skb->next = nskb;
389 curr_skb = nskb;
390 head_skb->truesize += nskb->truesize;
391 num_skb_frags = 0;
392 }
Michael Daltonab7db912014-01-16 22:23:27 -0800393 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
Michael Dalton2613af02013-10-28 15:44:18 -0700394 if (curr_skb != head_skb) {
395 head_skb->data_len += len;
396 head_skb->len += len;
Michael Daltonfb518792014-01-16 22:23:26 -0800397 head_skb->truesize += truesize;
Michael Dalton2613af02013-10-28 15:44:18 -0700398 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200399 offset = buf - page_address(page);
Jason Wangba275242013-11-01 14:07:48 +0800400 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
401 put_page(page);
402 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
Michael Daltonfb518792014-01-16 22:23:26 -0800403 len, truesize);
Jason Wangba275242013-11-01 14:07:48 +0800404 } else {
405 skb_add_rx_frag(curr_skb, num_skb_frags, page,
Michael Daltonfb518792014-01-16 22:23:26 -0800406 offset, len, truesize);
Jason Wangba275242013-11-01 14:07:48 +0800407 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200408 }
409
Johannes Berg5377d7582015-08-19 09:48:40 +0200410 ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200411 return head_skb;
412
413err_skb:
414 put_page(page);
415 while (--num_buf) {
Michael Daltonab7db912014-01-16 22:23:27 -0800416 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
417 if (unlikely(!ctx)) {
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200418 pr_debug("%s: rx error: %d buffers missing\n",
419 dev->name, num_buf);
420 dev->stats.rx_length_errors++;
421 break;
422 }
Michael Daltonab7db912014-01-16 22:23:27 -0800423 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200424 put_page(page);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000425 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200426err_buf:
427 dev->stats.rx_dropped++;
428 dev_kfree_skb(head_skb);
429 return NULL;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000430}
431
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300432static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
433 void *buf, unsigned int len)
Rusty Russell296f96f2007-10-22 11:03:37 +1000434{
Jason Wange9d74172012-12-07 07:04:55 +0000435 struct net_device *dev = vi->dev;
Eric Dumazet58472a72012-02-13 06:53:41 +0000436 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000437 struct sk_buff *skb;
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300438 struct virtio_net_hdr_mrg_rxbuf *hdr;
Rusty Russell296f96f2007-10-22 11:03:37 +1000439
Michael S. Tsirkinbcff3162014-10-24 00:22:11 +0300440 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000441 pr_debug("%s: short packet %i\n", dev->name, len);
442 dev->stats.rx_length_errors++;
Michael Daltonab7db912014-01-16 22:23:27 -0800443 if (vi->mergeable_rx_bufs) {
444 unsigned long ctx = (unsigned long)buf;
445 void *base = mergeable_ctx_to_buf_address(ctx);
446 put_page(virt_to_head_page(base));
447 } else if (vi->big_packets) {
Michael Dalton98bfd232013-12-05 13:14:05 -0800448 give_pages(rq, buf);
Michael Daltonab7db912014-01-16 22:23:27 -0800449 } else {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000450 dev_kfree_skb(buf);
Michael Daltonab7db912014-01-16 22:23:27 -0800451 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000452 return;
Rusty Russell296f96f2007-10-22 11:03:37 +1000453 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000454
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200455 if (vi->mergeable_rx_bufs)
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +0200456 skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len);
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200457 else if (vi->big_packets)
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300458 skb = receive_big(dev, vi, rq, buf, len);
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200459 else
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300460 skb = receive_small(vi, buf, len);
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200461
462 if (unlikely(!skb))
463 return;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800464
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000465 hdr = skb_vnet_hdr(skb);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000466
Eric Dumazet83a27052012-06-05 22:35:24 +0000467 u64_stats_update_begin(&stats->rx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000468 stats->rx_bytes += skb->len;
469 stats->rx_packets++;
Eric Dumazet83a27052012-06-05 22:35:24 +0000470 u64_stats_update_end(&stats->rx_syncp);
Rusty Russell296f96f2007-10-22 11:03:37 +1000471
Rusty Russellb3f24692009-09-24 09:59:19 -0600472 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000473 pr_debug("Needs csum!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600474 if (!skb_partial_csum_set(skb,
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +0200475 virtio16_to_cpu(vi->vdev, hdr->hdr.csum_start),
476 virtio16_to_cpu(vi->vdev, hdr->hdr.csum_offset)))
Rusty Russell296f96f2007-10-22 11:03:37 +1000477 goto frame_err;
Jason Wang10a8d942011-06-10 00:56:17 +0000478 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
479 skb->ip_summed = CHECKSUM_UNNECESSARY;
Rusty Russell296f96f2007-10-22 11:03:37 +1000480 }
481
Mark McLoughlin23cde762008-06-08 20:49:00 +1000482 skb->protocol = eth_type_trans(skb, dev);
483 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
484 ntohs(skb->protocol), skb->len, skb->pkt_type);
485
Rusty Russellb3f24692009-09-24 09:59:19 -0600486 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000487 pr_debug("GSO!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600488 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000489 case VIRTIO_NET_HDR_GSO_TCPV4:
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000490 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
Rusty Russell296f96f2007-10-22 11:03:37 +1000491 break;
Rusty Russell296f96f2007-10-22 11:03:37 +1000492 case VIRTIO_NET_HDR_GSO_UDP:
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000493 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Rusty Russell296f96f2007-10-22 11:03:37 +1000494 break;
495 case VIRTIO_NET_HDR_GSO_TCPV6:
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000496 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
Rusty Russell296f96f2007-10-22 11:03:37 +1000497 break;
498 default:
Amerigo Wangbe443892012-11-08 17:47:28 +0000499 net_warn_ratelimited("%s: bad gso type %u.\n",
500 dev->name, hdr->hdr.gso_type);
Rusty Russell296f96f2007-10-22 11:03:37 +1000501 goto frame_err;
502 }
503
Rusty Russellb3f24692009-09-24 09:59:19 -0600504 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000505 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
Rusty Russell34a48572008-02-04 23:50:02 -0500506
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +0200507 skb_shinfo(skb)->gso_size = virtio16_to_cpu(vi->vdev,
508 hdr->hdr.gso_size);
Rusty Russell296f96f2007-10-22 11:03:37 +1000509 if (skb_shinfo(skb)->gso_size == 0) {
Amerigo Wangbe443892012-11-08 17:47:28 +0000510 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
Rusty Russell296f96f2007-10-22 11:03:37 +1000511 goto frame_err;
512 }
513
514 /* Header must be checked, and gso_segs computed. */
515 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
516 skb_shinfo(skb)->gso_segs = 0;
517 }
518
Jason Wang91815632014-07-23 16:33:55 +0800519 skb_mark_napi_id(skb, &rq->napi);
520
Eric Dumazet0fbd0502015-07-31 18:25:17 +0200521 napi_gro_receive(&rq->napi, skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000522 return;
523
524frame_err:
525 dev->stats.rx_frame_errors++;
Rusty Russell296f96f2007-10-22 11:03:37 +1000526 dev_kfree_skb(skb);
527}
528
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300529static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
530 gfp_t gfp)
Rusty Russell296f96f2007-10-22 11:03:37 +1000531{
532 struct sk_buff *skb;
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300533 struct virtio_net_hdr_mrg_rxbuf *hdr;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000534 int err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000535
Michael Dalton5061de32013-11-14 10:41:04 -0800536 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000537 if (unlikely(!skb))
538 return -ENOMEM;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800539
Michael Dalton5061de32013-11-14 10:41:04 -0800540 skb_put(skb, GOOD_PACKET_LEN);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000541
542 hdr = skb_vnet_hdr(skb);
Jason Wang547c8902015-08-27 14:53:06 +0800543 sg_init_table(rq->sg, 2);
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300544 sg_set_buf(rq->sg, hdr, vi->hdr_len);
Jason Wange9d74172012-12-07 07:04:55 +0000545 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000546
Rusty Russell9dc7b9e2013-03-20 15:44:28 +1030547 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000548 if (err < 0)
549 dev_kfree_skb(skb);
550
551 return err;
552}
553
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300554static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
555 gfp_t gfp)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000556{
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000557 struct page *first, *list = NULL;
558 char *p;
559 int i, err, offset;
560
Rusty Russella5835442014-09-11 10:17:36 +0930561 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
562
Jason Wange9d74172012-12-07 07:04:55 +0000563 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000564 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
Jason Wange9d74172012-12-07 07:04:55 +0000565 first = get_a_page(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000566 if (!first) {
567 if (list)
Jason Wange9d74172012-12-07 07:04:55 +0000568 give_pages(rq, list);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000569 return -ENOMEM;
Rusty Russell3161e452009-08-26 12:22:32 -0700570 }
Jason Wange9d74172012-12-07 07:04:55 +0000571 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
Rusty Russell296f96f2007-10-22 11:03:37 +1000572
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000573 /* chain new page in list head to match sg */
574 first->private = (unsigned long)list;
575 list = first;
576 }
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800577
Jason Wange9d74172012-12-07 07:04:55 +0000578 first = get_a_page(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000579 if (!first) {
Jason Wange9d74172012-12-07 07:04:55 +0000580 give_pages(rq, list);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000581 return -ENOMEM;
582 }
583 p = page_address(first);
Herbert Xu97402b92008-04-18 11:24:27 +0800584
Jason Wange9d74172012-12-07 07:04:55 +0000585 /* rq->sg[0], rq->sg[1] share the same page */
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300586 /* a separated rq->sg[0] for header - required in case !any_header_sg */
587 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
Herbert Xu97402b92008-04-18 11:24:27 +0800588
Jason Wange9d74172012-12-07 07:04:55 +0000589 /* rq->sg[1] for data packet, from offset */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000590 offset = sizeof(struct padded_vnet_hdr);
Jason Wange9d74172012-12-07 07:04:55 +0000591 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
Herbert Xu97402b92008-04-18 11:24:27 +0800592
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000593 /* chain first in list head */
594 first->private = (unsigned long)list;
Rusty Russell9dc7b9e2013-03-20 15:44:28 +1030595 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
596 first, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000597 if (err < 0)
Jason Wange9d74172012-12-07 07:04:55 +0000598 give_pages(rq, first);
Herbert Xu97402b92008-04-18 11:24:27 +0800599
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000600 return err;
601}
Herbert Xu97402b92008-04-18 11:24:27 +0800602
Johannes Berg5377d7582015-08-19 09:48:40 +0200603static unsigned int get_mergeable_buf_len(struct ewma_pkt_len *avg_pkt_len)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000604{
Michael Daltonab7db912014-01-16 22:23:27 -0800605 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
Michael Daltonfbf28d72014-01-16 22:23:30 -0800606 unsigned int len;
607
Johannes Berg5377d7582015-08-19 09:48:40 +0200608 len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
Michael Daltonfbf28d72014-01-16 22:23:30 -0800609 GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
610 return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
611}
612
613static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
614{
Michael Daltonfb518792014-01-16 22:23:26 -0800615 struct page_frag *alloc_frag = &rq->alloc_frag;
616 char *buf;
Michael Daltonab7db912014-01-16 22:23:27 -0800617 unsigned long ctx;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000618 int err;
Michael Daltonfb518792014-01-16 22:23:26 -0800619 unsigned int len, hole;
Rusty Russell296f96f2007-10-22 11:03:37 +1000620
Michael Daltonfbf28d72014-01-16 22:23:30 -0800621 len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
Michael Daltonab7db912014-01-16 22:23:27 -0800622 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000623 return -ENOMEM;
Michael Daltonab7db912014-01-16 22:23:27 -0800624
Michael Daltonfb518792014-01-16 22:23:26 -0800625 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
Michael Daltonab7db912014-01-16 22:23:27 -0800626 ctx = mergeable_buf_to_ctx(buf, len);
Michael Daltonfb518792014-01-16 22:23:26 -0800627 get_page(alloc_frag->page);
Michael Daltonfb518792014-01-16 22:23:26 -0800628 alloc_frag->offset += len;
629 hole = alloc_frag->size - alloc_frag->offset;
Michael Daltonab7db912014-01-16 22:23:27 -0800630 if (hole < len) {
631 /* To avoid internal fragmentation, if there is very likely not
632 * enough space for another buffer, add the remaining space to
633 * the current buffer. This extra space is not included in
634 * the truesize stored in ctx.
635 */
Michael Daltonfb518792014-01-16 22:23:26 -0800636 len += hole;
637 alloc_frag->offset += hole;
638 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000639
Michael Daltonfb518792014-01-16 22:23:26 -0800640 sg_init_one(rq->sg, buf, len);
Michael Daltonab7db912014-01-16 22:23:27 -0800641 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000642 if (err < 0)
Michael Dalton2613af02013-10-28 15:44:18 -0700643 put_page(virt_to_head_page(buf));
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000644
645 return err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000646}
647
Rusty Russellb2baed62011-12-29 00:42:38 +0000648/*
649 * Returns false if we couldn't fill entirely (OOM).
650 *
651 * Normally run in the receive path, but can also be run from ndo_open
652 * before we're receiving packets, or from refill_work which is
653 * careful to disable receiving (using napi_disable).
654 */
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300655static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
656 gfp_t gfp)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800657{
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800658 int err;
Michael S. Tsirkin1788f4952010-07-02 16:32:55 +0000659 bool oom;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800660
Michael Daltonfb518792014-01-16 22:23:26 -0800661 gfp |= __GFP_COLD;
Amit Shah0aea51c2009-08-26 14:58:28 +0530662 do {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000663 if (vi->mergeable_rx_bufs)
Jason Wange9d74172012-12-07 07:04:55 +0000664 err = add_recvbuf_mergeable(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000665 else if (vi->big_packets)
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300666 err = add_recvbuf_big(vi, rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000667 else
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300668 err = add_recvbuf_small(vi, rq, gfp);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800669
Michael S. Tsirkin1788f4952010-07-02 16:32:55 +0000670 oom = err == -ENOMEM;
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030671 if (err)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800672 break;
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800673 } while (rq->vq->num_free);
Jason Wang681daee22014-03-26 13:03:00 +0800674 virtqueue_kick(rq->vq);
Rusty Russell3161e452009-08-26 12:22:32 -0700675 return !oom;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800676}
677
Rusty Russell18445c42008-02-04 23:49:57 -0500678static void skb_recv_done(struct virtqueue *rvq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000679{
680 struct virtnet_info *vi = rvq->vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +0000681 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
Jason Wange9d74172012-12-07 07:04:55 +0000682
Rusty Russell18445c42008-02-04 23:49:57 -0500683 /* Schedule NAPI, Suppress further interrupts if successful. */
Jason Wange9d74172012-12-07 07:04:55 +0000684 if (napi_schedule_prep(&rq->napi)) {
Michael S. Tsirkin1915a7122010-04-12 16:19:04 +0300685 virtqueue_disable_cb(rvq);
Jason Wange9d74172012-12-07 07:04:55 +0000686 __napi_schedule(&rq->napi);
Rusty Russell18445c42008-02-04 23:49:57 -0500687 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000688}
689
Jason Wange9d74172012-12-07 07:04:55 +0000690static void virtnet_napi_enable(struct receive_queue *rq)
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800691{
Jason Wange9d74172012-12-07 07:04:55 +0000692 napi_enable(&rq->napi);
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800693
694 /* If all buffers were filled by other side before we napi_enabled, we
695 * won't get another interrupt, so process any outstanding packets
696 * now. virtnet_poll wants re-enable the queue, so we disable here.
697 * We synchronize against interrupts via NAPI_STATE_SCHED */
Jason Wange9d74172012-12-07 07:04:55 +0000698 if (napi_schedule_prep(&rq->napi)) {
699 virtqueue_disable_cb(rq->vq);
Michael S. Tsirkinec13ee82012-05-16 10:57:12 +0300700 local_bh_disable();
Jason Wange9d74172012-12-07 07:04:55 +0000701 __napi_schedule(&rq->napi);
Michael S. Tsirkinec13ee82012-05-16 10:57:12 +0300702 local_bh_enable();
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800703 }
704}
705
Rusty Russell3161e452009-08-26 12:22:32 -0700706static void refill_work(struct work_struct *work)
707{
Jason Wange9d74172012-12-07 07:04:55 +0000708 struct virtnet_info *vi =
709 container_of(work, struct virtnet_info, refill.work);
Rusty Russell3161e452009-08-26 12:22:32 -0700710 bool still_empty;
Jason Wang986a4f42012-12-07 07:04:56 +0000711 int i;
Rusty Russell3161e452009-08-26 12:22:32 -0700712
Sasha Levin55257d72013-04-29 12:00:08 +0930713 for (i = 0; i < vi->curr_queue_pairs; i++) {
Jason Wang986a4f42012-12-07 07:04:56 +0000714 struct receive_queue *rq = &vi->rq[i];
Rusty Russell3161e452009-08-26 12:22:32 -0700715
Jason Wang986a4f42012-12-07 07:04:56 +0000716 napi_disable(&rq->napi);
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300717 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
Jason Wang986a4f42012-12-07 07:04:56 +0000718 virtnet_napi_enable(rq);
719
720 /* In theory, this can happen: if we don't get any buffers in
721 * we will *never* try to fill again.
722 */
723 if (still_empty)
724 schedule_delayed_work(&vi->refill, HZ/2);
725 }
Rusty Russell3161e452009-08-26 12:22:32 -0700726}
727
Jason Wang2ffa7592014-07-23 16:33:54 +0800728static int virtnet_receive(struct receive_queue *rq, int budget)
Rusty Russell296f96f2007-10-22 11:03:37 +1000729{
Jason Wange9d74172012-12-07 07:04:55 +0000730 struct virtnet_info *vi = rq->vq->vdev->priv;
Jason Wang2ffa7592014-07-23 16:33:54 +0800731 unsigned int len, received = 0;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000732 void *buf;
Rusty Russell296f96f2007-10-22 11:03:37 +1000733
Rusty Russell296f96f2007-10-22 11:03:37 +1000734 while (received < budget &&
Jason Wange9d74172012-12-07 07:04:55 +0000735 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300736 receive_buf(vi, rq, buf, len);
Rusty Russell296f96f2007-10-22 11:03:37 +1000737 received++;
738 }
739
Jason Wangbe121f42014-01-16 14:45:24 +0800740 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300741 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
Tejun Heo3b07e9c2012-08-20 14:51:24 -0700742 schedule_delayed_work(&vi->refill, 0);
Rusty Russell3161e452009-08-26 12:22:32 -0700743 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000744
Jason Wang2ffa7592014-07-23 16:33:54 +0800745 return received;
746}
747
748static int virtnet_poll(struct napi_struct *napi, int budget)
749{
750 struct receive_queue *rq =
751 container_of(napi, struct receive_queue, napi);
Li RongQingfaadb052015-03-26 15:39:45 +0800752 unsigned int r, received;
Jason Wang2ffa7592014-07-23 16:33:54 +0800753
Li RongQingfaadb052015-03-26 15:39:45 +0800754 received = virtnet_receive(rq, budget);
Jason Wang2ffa7592014-07-23 16:33:54 +0800755
Rusty Russell8329d982007-11-19 11:20:43 -0500756 /* Out of packets? */
757 if (received < budget) {
Michael S. Tsirkincbdadbb2013-07-09 08:13:04 +0300758 r = virtqueue_enable_cb_prepare(rq->vq);
Eric Dumazet0fbd0502015-07-31 18:25:17 +0200759 napi_complete_done(napi, received);
Michael S. Tsirkincbdadbb2013-07-09 08:13:04 +0300760 if (unlikely(virtqueue_poll(rq->vq, r)) &&
Joe Perches8e95a202009-12-03 07:58:21 +0000761 napi_schedule_prep(napi)) {
Jason Wange9d74172012-12-07 07:04:55 +0000762 virtqueue_disable_cb(rq->vq);
Ben Hutchings288379f2009-01-19 16:43:59 -0800763 __napi_schedule(napi);
Christian Borntraeger4265f162008-03-14 14:17:05 +0100764 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000765 }
766
767 return received;
768}
769
Jason Wang91815632014-07-23 16:33:55 +0800770#ifdef CONFIG_NET_RX_BUSY_POLL
771/* must be called with local_bh_disable()d */
772static int virtnet_busy_poll(struct napi_struct *napi)
773{
774 struct receive_queue *rq =
775 container_of(napi, struct receive_queue, napi);
776 struct virtnet_info *vi = rq->vq->vdev->priv;
777 int r, received = 0, budget = 4;
778
779 if (!(vi->status & VIRTIO_NET_S_LINK_UP))
780 return LL_FLUSH_FAILED;
781
782 if (!napi_schedule_prep(napi))
783 return LL_FLUSH_BUSY;
784
785 virtqueue_disable_cb(rq->vq);
786
787again:
788 received += virtnet_receive(rq, budget);
789
790 r = virtqueue_enable_cb_prepare(rq->vq);
791 clear_bit(NAPI_STATE_SCHED, &napi->state);
792 if (unlikely(virtqueue_poll(rq->vq, r)) &&
793 napi_schedule_prep(napi)) {
794 virtqueue_disable_cb(rq->vq);
795 if (received < budget) {
796 budget -= received;
797 goto again;
798 } else {
799 __napi_schedule(napi);
800 }
801 }
802
803 return received;
804}
805#endif /* CONFIG_NET_RX_BUSY_POLL */
806
Jason Wang986a4f42012-12-07 07:04:56 +0000807static int virtnet_open(struct net_device *dev)
808{
809 struct virtnet_info *vi = netdev_priv(dev);
810 int i;
811
Jason Wange4166622013-05-21 20:03:58 +0000812 for (i = 0; i < vi->max_queue_pairs; i++) {
813 if (i < vi->curr_queue_pairs)
814 /* Make sure we have some buffers: if oom use wq. */
Michael S. Tsirkin946fa562014-10-24 00:12:10 +0300815 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
Jason Wange4166622013-05-21 20:03:58 +0000816 schedule_delayed_work(&vi->refill, 0);
Jason Wang986a4f42012-12-07 07:04:56 +0000817 virtnet_napi_enable(&vi->rq[i]);
818 }
819
820 return 0;
821}
822
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800823static void free_old_xmit_skbs(struct send_queue *sq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000824{
825 struct sk_buff *skb;
Michael S. Tsirkin6ee57bc2012-10-16 23:56:14 +1030826 unsigned int len;
Jason Wange9d74172012-12-07 07:04:55 +0000827 struct virtnet_info *vi = sq->vq->vdev->priv;
Eric Dumazet58472a72012-02-13 06:53:41 +0000828 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
Rusty Russell296f96f2007-10-22 11:03:37 +1000829
Jason Wange9d74172012-12-07 07:04:55 +0000830 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000831 pr_debug("Sent skb %p\n", skb);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000832
Eric Dumazet83a27052012-06-05 22:35:24 +0000833 u64_stats_update_begin(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000834 stats->tx_bytes += skb->len;
835 stats->tx_packets++;
Eric Dumazet83a27052012-06-05 22:35:24 +0000836 u64_stats_update_end(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000837
Eric Dumazeted79bab2009-10-14 14:36:43 +0000838 dev_kfree_skb_any(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000839 }
840}
841
Jason Wange9d74172012-12-07 07:04:55 +0000842static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +1000843{
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300844 struct virtio_net_hdr_mrg_rxbuf *hdr;
Rusty Russell296f96f2007-10-22 11:03:37 +1000845 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
Jason Wange9d74172012-12-07 07:04:55 +0000846 struct virtnet_info *vi = sq->vq->vdev->priv;
Michael S. Tsirkin7bedc7d2012-10-16 23:56:14 +1030847 unsigned num_sg;
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300848 unsigned hdr_len = vi->hdr_len;
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930849 bool can_push;
Rusty Russell296f96f2007-10-22 11:03:37 +1000850
Johannes Berge1749612008-10-27 15:59:26 -0700851 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930852
853 can_push = vi->any_header_sg &&
854 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
855 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
856 /* Even if we can, don't push here yet as this would skew
857 * csum_start offset below. */
858 if (can_push)
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300859 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930860 else
861 hdr = skb_vnet_hdr(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000862
Rusty Russell296f96f2007-10-22 11:03:37 +1000863 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Rusty Russellb3f24692009-09-24 09:59:19 -0600864 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +0200865 hdr->hdr.csum_start = cpu_to_virtio16(vi->vdev,
866 skb_checksum_start_offset(skb));
867 hdr->hdr.csum_offset = cpu_to_virtio16(vi->vdev,
868 skb->csum_offset);
Rusty Russell296f96f2007-10-22 11:03:37 +1000869 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600870 hdr->hdr.flags = 0;
871 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000872 }
873
874 if (skb_is_gso(skb)) {
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +0200875 hdr->hdr.hdr_len = cpu_to_virtio16(vi->vdev, skb_headlen(skb));
876 hdr->hdr.gso_size = cpu_to_virtio16(vi->vdev,
877 skb_shinfo(skb)->gso_size);
Rusty Russell34a48572008-02-04 23:50:02 -0500878 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
Rusty Russellb3f24692009-09-24 09:59:19 -0600879 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
Rusty Russell296f96f2007-10-22 11:03:37 +1000880 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
Rusty Russellb3f24692009-09-24 09:59:19 -0600881 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
Vlad Yaseviche3e3c422015-02-03 16:36:17 -0500882 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
883 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
Rusty Russell296f96f2007-10-22 11:03:37 +1000884 else
885 BUG();
Rusty Russell34a48572008-02-04 23:50:02 -0500886 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
Rusty Russellb3f24692009-09-24 09:59:19 -0600887 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
Rusty Russell296f96f2007-10-22 11:03:37 +1000888 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600889 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
890 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000891 }
892
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800893 if (vi->mergeable_rx_bufs)
Michael S. Tsirkin012873d2014-10-24 16:55:57 +0300894 hdr->num_buffers = 0;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800895
Jason Wang547c8902015-08-27 14:53:06 +0800896 sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930897 if (can_push) {
898 __skb_push(skb, hdr_len);
899 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
900 /* Pull header back to avoid skew in tx bytes calculations. */
901 __skb_pull(skb, hdr_len);
902 } else {
903 sg_set_buf(sq->sg, hdr, hdr_len);
904 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
905 }
Rusty Russell9dc7b9e2013-03-20 15:44:28 +1030906 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
Rusty Russell11a3a152008-05-26 17:48:13 +1000907}
908
Stephen Hemminger424efe92009-08-31 19:50:51 +0000909static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
Rusty Russell99ffc692008-05-02 21:50:46 -0500910{
911 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +0000912 int qnum = skb_get_queue_mapping(skb);
913 struct send_queue *sq = &vi->sq[qnum];
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030914 int err;
Michael S. Tsirkin4b7fd2e62014-10-15 16:23:28 +0300915 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
916 bool kick = !skb->xmit_more;
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500917
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500918 /* Free up any pending old buffers before queueing new ones. */
Jason Wange9d74172012-12-07 07:04:55 +0000919 free_old_xmit_skbs(sq);
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500920
Jacob Keller074c3582014-06-25 02:37:13 +0000921 /* timestamp packet in software */
922 skb_tx_timestamp(skb);
923
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700924 /* Try to transmit */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800925 err = xmit_skb(sq, skb);
Rusty Russell48925e32009-09-24 09:59:20 -0600926
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030927 /* This should not happen! */
Jason Wang681daee22014-03-26 13:03:00 +0800928 if (unlikely(err)) {
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030929 dev->stats.tx_fifo_errors++;
930 if (net_ratelimit())
931 dev_warn(&dev->dev,
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800932 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
Rusty Russell58eba972010-07-02 16:34:01 +0000933 dev->stats.tx_dropped++;
Eric W. Biederman85e94522014-03-15 18:43:33 -0700934 dev_kfree_skb_any(skb);
Rusty Russell58eba972010-07-02 16:34:01 +0000935 return NETDEV_TX_OK;
Rusty Russell99ffc692008-05-02 21:50:46 -0500936 }
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700937
Rusty Russell48925e32009-09-24 09:59:20 -0600938 /* Don't wait up for transmitted skbs to be freed. */
939 skb_orphan(skb);
940 nf_reset(skb);
Rusty Russell99ffc692008-05-02 21:50:46 -0500941
Michael S. Tsirkin60302ff2015-04-02 13:05:47 +0200942 /* If running out of space, stop queue to avoid getting packets that we
943 * are then unable to transmit.
944 * An alternative would be to force queuing layer to requeue the skb by
945 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
946 * returned in a normal path of operation: it means that driver is not
947 * maintaining the TX queue stop/start state properly, and causes
948 * the stack to do a non-trivial amount of useless work.
949 * Since most packets only take 1 or 2 ring slots, stopping the queue
950 * early means 16 slots are typically wasted.
stephen hemmingerd631b942015-03-24 16:22:07 -0700951 */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800952 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
Jason Wang986a4f42012-12-07 07:04:56 +0000953 netif_stop_subqueue(dev, qnum);
Jason Wange9d74172012-12-07 07:04:55 +0000954 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
Rusty Russell48925e32009-09-24 09:59:20 -0600955 /* More just got used, free them then recheck. */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800956 free_old_xmit_skbs(sq);
957 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
Jason Wang986a4f42012-12-07 07:04:56 +0000958 netif_start_subqueue(dev, qnum);
Jason Wange9d74172012-12-07 07:04:55 +0000959 virtqueue_disable_cb(sq->vq);
Rusty Russell48925e32009-09-24 09:59:20 -0600960 }
961 }
Rusty Russell99ffc692008-05-02 21:50:46 -0500962 }
Rusty Russell48925e32009-09-24 09:59:20 -0600963
Michael S. Tsirkin4b7fd2e62014-10-15 16:23:28 +0300964 if (kick || netif_xmit_stopped(txq))
David S. Miller0b725a22014-08-25 15:51:53 -0700965 virtqueue_kick(sq->vq);
966
Rusty Russell48925e32009-09-24 09:59:20 -0600967 return NETDEV_TX_OK;
Rusty Russell296f96f2007-10-22 11:03:37 +1000968}
969
Amos Kong40cbfc32013-01-21 01:17:21 +0000970/*
971 * Send command via the control virtqueue and check status. Commands
972 * supported by the hypervisor, as indicated by feature bits, should
stephen hemminger788a8b62013-12-09 16:18:45 -0800973 * never fail unless improperly formatted.
Amos Kong40cbfc32013-01-21 01:17:21 +0000974 */
975static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
stephen hemmingerd24bae32013-12-09 16:17:40 -0800976 struct scatterlist *out)
Amos Kong40cbfc32013-01-21 01:17:21 +0000977{
Rusty Russellf7bc9592013-03-20 15:44:28 +1030978 struct scatterlist *sgs[4], hdr, stat;
Amos Kong40cbfc32013-01-21 01:17:21 +0000979 struct virtio_net_ctrl_hdr ctrl;
980 virtio_net_ctrl_ack status = ~0;
stephen hemmingerd24bae32013-12-09 16:17:40 -0800981 unsigned out_num = 0, tmp;
Amos Kong40cbfc32013-01-21 01:17:21 +0000982
983 /* Caller should know better */
Rusty Russellf7bc9592013-03-20 15:44:28 +1030984 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
Amos Kong40cbfc32013-01-21 01:17:21 +0000985
986 ctrl.class = class;
987 ctrl.cmd = cmd;
Rusty Russellf7bc9592013-03-20 15:44:28 +1030988 /* Add header */
989 sg_init_one(&hdr, &ctrl, sizeof(ctrl));
990 sgs[out_num++] = &hdr;
Amos Kong40cbfc32013-01-21 01:17:21 +0000991
Rusty Russellf7bc9592013-03-20 15:44:28 +1030992 if (out)
993 sgs[out_num++] = out;
Amos Kong40cbfc32013-01-21 01:17:21 +0000994
Rusty Russellf7bc9592013-03-20 15:44:28 +1030995 /* Add return status. */
996 sg_init_one(&stat, &status, sizeof(status));
stephen hemmingerd24bae32013-12-09 16:17:40 -0800997 sgs[out_num] = &stat;
Amos Kong40cbfc32013-01-21 01:17:21 +0000998
stephen hemmingerd24bae32013-12-09 16:17:40 -0800999 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
Rusty Russella7c58142014-03-13 11:23:39 +10301000 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
Amos Kong40cbfc32013-01-21 01:17:21 +00001001
Heinz Graalfs67975902013-10-29 09:40:02 +10301002 if (unlikely(!virtqueue_kick(vi->cvq)))
1003 return status == VIRTIO_NET_OK;
Amos Kong40cbfc32013-01-21 01:17:21 +00001004
1005 /* Spin for a response, the kick causes an ioport write, trapping
1006 * into the hypervisor, so the request should be handled immediately.
1007 */
Heinz Graalfs047b9b92013-10-29 09:40:47 +10301008 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1009 !virtqueue_is_broken(vi->cvq))
Amos Kong40cbfc32013-01-21 01:17:21 +00001010 cpu_relax();
1011
1012 return status == VIRTIO_NET_OK;
1013}
1014
Alex Williamson9c46f6d2009-02-04 16:36:34 -08001015static int virtnet_set_mac_address(struct net_device *dev, void *p)
1016{
1017 struct virtnet_info *vi = netdev_priv(dev);
1018 struct virtio_device *vdev = vi->vdev;
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +00001019 int ret;
Amos Kong7e58d5a2013-01-21 01:17:23 +00001020 struct sockaddr *addr = p;
1021 struct scatterlist sg;
Alex Williamson9c46f6d2009-02-04 16:36:34 -08001022
Amos Kong7e58d5a2013-01-21 01:17:23 +00001023 ret = eth_prepare_mac_addr_change(dev, p);
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +00001024 if (ret)
1025 return ret;
Alex Williamson9c46f6d2009-02-04 16:36:34 -08001026
Amos Kong7e58d5a2013-01-21 01:17:23 +00001027 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1028 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1029 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001030 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
Amos Kong7e58d5a2013-01-21 01:17:23 +00001031 dev_warn(&vdev->dev,
1032 "Failed to set mac address by vq command.\n");
1033 return -EINVAL;
1034 }
Michael S. Tsirkin7e93a022014-11-26 15:58:28 +02001035 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1036 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
Rusty Russell855e0c52013-10-14 18:11:51 +10301037 unsigned int i;
1038
1039 /* Naturally, this has an atomicity problem. */
1040 for (i = 0; i < dev->addr_len; i++)
1041 virtio_cwrite8(vdev,
1042 offsetof(struct virtio_net_config, mac) +
1043 i, addr->sa_data[i]);
Amos Kong7e58d5a2013-01-21 01:17:23 +00001044 }
1045
1046 eth_commit_mac_addr_change(dev, p);
Alex Williamson9c46f6d2009-02-04 16:36:34 -08001047
1048 return 0;
1049}
1050
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001051static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
1052 struct rtnl_link_stats64 *tot)
1053{
1054 struct virtnet_info *vi = netdev_priv(dev);
1055 int cpu;
1056 unsigned int start;
1057
1058 for_each_possible_cpu(cpu) {
Eric Dumazet58472a72012-02-13 06:53:41 +00001059 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001060 u64 tpackets, tbytes, rpackets, rbytes;
1061
1062 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -07001063 start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001064 tpackets = stats->tx_packets;
1065 tbytes = stats->tx_bytes;
Eric W. Biederman57a77442014-03-13 21:26:42 -07001066 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
Eric Dumazet83a27052012-06-05 22:35:24 +00001067
1068 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -07001069 start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001070 rpackets = stats->rx_packets;
1071 rbytes = stats->rx_bytes;
Eric W. Biederman57a77442014-03-13 21:26:42 -07001072 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001073
1074 tot->rx_packets += rpackets;
1075 tot->tx_packets += tpackets;
1076 tot->rx_bytes += rbytes;
1077 tot->tx_bytes += tbytes;
1078 }
1079
1080 tot->tx_dropped = dev->stats.tx_dropped;
Rick Jones021ac8d2011-11-21 09:28:17 +00001081 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001082 tot->rx_dropped = dev->stats.rx_dropped;
1083 tot->rx_length_errors = dev->stats.rx_length_errors;
1084 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1085
1086 return tot;
1087}
1088
Amit Shahda74e892008-02-29 16:24:50 +05301089#ifdef CONFIG_NET_POLL_CONTROLLER
1090static void virtnet_netpoll(struct net_device *dev)
1091{
1092 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001093 int i;
Amit Shahda74e892008-02-29 16:24:50 +05301094
Jason Wang986a4f42012-12-07 07:04:56 +00001095 for (i = 0; i < vi->curr_queue_pairs; i++)
1096 napi_schedule(&vi->rq[i].napi);
Amit Shahda74e892008-02-29 16:24:50 +05301097}
1098#endif
1099
Jason Wang586d17c2012-04-11 20:43:52 +00001100static void virtnet_ack_link_announce(struct virtnet_info *vi)
1101{
1102 rtnl_lock();
1103 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001104 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
Jason Wang586d17c2012-04-11 20:43:52 +00001105 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1106 rtnl_unlock();
1107}
1108
Jason Wang986a4f42012-12-07 07:04:56 +00001109static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1110{
1111 struct scatterlist sg;
1112 struct virtio_net_ctrl_mq s;
1113 struct net_device *dev = vi->dev;
1114
1115 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1116 return 0;
1117
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +02001118 s.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
Jason Wang986a4f42012-12-07 07:04:56 +00001119 sg_init_one(&sg, &s, sizeof(s));
1120
1121 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001122 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
Jason Wang986a4f42012-12-07 07:04:56 +00001123 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1124 queue_pairs);
1125 return -EINVAL;
Sasha Levin55257d72013-04-29 12:00:08 +09301126 } else {
Jason Wang986a4f42012-12-07 07:04:56 +00001127 vi->curr_queue_pairs = queue_pairs;
Jason Wang35ed1592013-10-15 11:18:59 +08001128 /* virtnet_open() will refill when device is going to up. */
1129 if (dev->flags & IFF_UP)
1130 schedule_delayed_work(&vi->refill, 0);
Sasha Levin55257d72013-04-29 12:00:08 +09301131 }
Jason Wang986a4f42012-12-07 07:04:56 +00001132
1133 return 0;
1134}
1135
Rusty Russell296f96f2007-10-22 11:03:37 +10001136static int virtnet_close(struct net_device *dev)
1137{
1138 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001139 int i;
Rusty Russell296f96f2007-10-22 11:03:37 +10001140
Rusty Russellb2baed62011-12-29 00:42:38 +00001141 /* Make sure refill_work doesn't re-enable napi! */
1142 cancel_delayed_work_sync(&vi->refill);
Jason Wang986a4f42012-12-07 07:04:56 +00001143
1144 for (i = 0; i < vi->max_queue_pairs; i++)
1145 napi_disable(&vi->rq[i].napi);
Rusty Russell296f96f2007-10-22 11:03:37 +10001146
Rusty Russell296f96f2007-10-22 11:03:37 +10001147 return 0;
1148}
1149
Alex Williamson2af76982009-02-04 09:02:40 +00001150static void virtnet_set_rx_mode(struct net_device *dev)
1151{
1152 struct virtnet_info *vi = netdev_priv(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001153 struct scatterlist sg[2];
Alex Williamson2af76982009-02-04 09:02:40 +00001154 u8 promisc, allmulti;
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001155 struct virtio_net_ctrl_mac *mac_data;
Jiri Pirkoccffad252009-05-22 23:22:17 +00001156 struct netdev_hw_addr *ha;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001157 int uc_count;
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001158 int mc_count;
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001159 void *buf;
1160 int i;
Alex Williamson2af76982009-02-04 09:02:40 +00001161
stephen hemminger788a8b62013-12-09 16:18:45 -08001162 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
Alex Williamson2af76982009-02-04 09:02:40 +00001163 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1164 return;
1165
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001166 promisc = ((dev->flags & IFF_PROMISC) != 0);
1167 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
Alex Williamson2af76982009-02-04 09:02:40 +00001168
Alex Williamson23e258e2009-05-01 17:27:56 +00001169 sg_init_one(sg, &promisc, sizeof(promisc));
Alex Williamson2af76982009-02-04 09:02:40 +00001170
1171 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001172 VIRTIO_NET_CTRL_RX_PROMISC, sg))
Alex Williamson2af76982009-02-04 09:02:40 +00001173 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1174 promisc ? "en" : "dis");
1175
Alex Williamson23e258e2009-05-01 17:27:56 +00001176 sg_init_one(sg, &allmulti, sizeof(allmulti));
Alex Williamson2af76982009-02-04 09:02:40 +00001177
1178 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001179 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
Alex Williamson2af76982009-02-04 09:02:40 +00001180 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1181 allmulti ? "en" : "dis");
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001182
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001183 uc_count = netdev_uc_count(dev);
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001184 mc_count = netdev_mc_count(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001185 /* MAC filter - use one buffer for both lists */
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001186 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1187 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1188 mac_data = buf;
Joe Perchese68ed8f2013-02-03 17:28:15 +00001189 if (!buf)
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001190 return;
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001191
Alex Williamson23e258e2009-05-01 17:27:56 +00001192 sg_init_table(sg, 2);
1193
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001194 /* Store the unicast list and count in the front of the buffer */
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +02001195 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
Jiri Pirkoccffad252009-05-22 23:22:17 +00001196 i = 0;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001197 netdev_for_each_uc_addr(ha, dev)
Jiri Pirkoccffad252009-05-22 23:22:17 +00001198 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001199
1200 sg_set_buf(&sg[0], mac_data,
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001201 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001202
1203 /* multicast list and count fill the end */
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001204 mac_data = (void *)&mac_data->macs[uc_count][0];
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001205
Michael S. Tsirkinfdd819b2014-10-07 16:39:48 +02001206 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
Jiri Pirko567ec872010-02-23 23:17:07 +00001207 i = 0;
Jiri Pirko22bedad32010-04-01 21:22:57 +00001208 netdev_for_each_mc_addr(ha, dev)
1209 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001210
1211 sg_set_buf(&sg[1], mac_data,
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001212 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001213
1214 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001215 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
Thomas Huth99e872a2013-11-29 10:02:19 +01001216 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001217
1218 kfree(buf);
Alex Williamson2af76982009-02-04 09:02:40 +00001219}
1220
Patrick McHardy80d5c362013-04-19 02:04:28 +00001221static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1222 __be16 proto, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +00001223{
1224 struct virtnet_info *vi = netdev_priv(dev);
1225 struct scatterlist sg;
1226
Alex Williamson23e258e2009-05-01 17:27:56 +00001227 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +00001228
1229 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001230 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
Alex Williamson0bde95692009-02-04 09:02:50 +00001231 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
Jiri Pirko8e586132011-12-08 19:52:37 -05001232 return 0;
Alex Williamson0bde95692009-02-04 09:02:50 +00001233}
1234
Patrick McHardy80d5c362013-04-19 02:04:28 +00001235static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1236 __be16 proto, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +00001237{
1238 struct virtnet_info *vi = netdev_priv(dev);
1239 struct scatterlist sg;
1240
Alex Williamson23e258e2009-05-01 17:27:56 +00001241 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +00001242
1243 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001244 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
Alex Williamson0bde95692009-02-04 09:02:50 +00001245 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
Jiri Pirko8e586132011-12-08 19:52:37 -05001246 return 0;
Alex Williamson0bde95692009-02-04 09:02:50 +00001247}
1248
Wanlong Gao8898c212013-01-24 23:51:30 +00001249static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
Jason Wang986a4f42012-12-07 07:04:56 +00001250{
1251 int i;
Wanlong Gao8898c212013-01-24 23:51:30 +00001252
1253 if (vi->affinity_hint_set) {
1254 for (i = 0; i < vi->max_queue_pairs; i++) {
1255 virtqueue_set_affinity(vi->rq[i].vq, -1);
1256 virtqueue_set_affinity(vi->sq[i].vq, -1);
1257 }
1258
1259 vi->affinity_hint_set = false;
1260 }
Wanlong Gao8898c212013-01-24 23:51:30 +00001261}
1262
1263static void virtnet_set_affinity(struct virtnet_info *vi)
Jason Wang986a4f42012-12-07 07:04:56 +00001264{
1265 int i;
Wanlong Gao47be2472013-01-24 23:51:29 +00001266 int cpu;
Jason Wang986a4f42012-12-07 07:04:56 +00001267
1268 /* In multiqueue mode, when the number of cpu is equal to the number of
1269 * queue pairs, we let the queue pairs to be private to one cpu by
1270 * setting the affinity hint to eliminate the contention.
1271 */
Wanlong Gao8898c212013-01-24 23:51:30 +00001272 if (vi->curr_queue_pairs == 1 ||
1273 vi->max_queue_pairs != num_online_cpus()) {
1274 virtnet_clean_affinity(vi, -1);
1275 return;
Jason Wang986a4f42012-12-07 07:04:56 +00001276 }
1277
Wanlong Gao8898c212013-01-24 23:51:30 +00001278 i = 0;
1279 for_each_online_cpu(cpu) {
Jason Wang986a4f42012-12-07 07:04:56 +00001280 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1281 virtqueue_set_affinity(vi->sq[i].vq, cpu);
Jason Wang9bb8ca82013-11-05 18:19:45 +08001282 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
Wanlong Gao8898c212013-01-24 23:51:30 +00001283 i++;
Jason Wang986a4f42012-12-07 07:04:56 +00001284 }
1285
Wanlong Gao8898c212013-01-24 23:51:30 +00001286 vi->affinity_hint_set = true;
Jason Wang986a4f42012-12-07 07:04:56 +00001287}
1288
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001289static int virtnet_cpu_callback(struct notifier_block *nfb,
1290 unsigned long action, void *hcpu)
1291{
1292 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1293
1294 switch(action & ~CPU_TASKS_FROZEN) {
1295 case CPU_ONLINE:
1296 case CPU_DOWN_FAILED:
1297 case CPU_DEAD:
1298 virtnet_set_affinity(vi);
1299 break;
1300 case CPU_DOWN_PREPARE:
1301 virtnet_clean_affinity(vi, (long)hcpu);
1302 break;
1303 default:
1304 break;
1305 }
Jason Wang3ab098d2013-10-15 11:18:58 +08001306
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001307 return NOTIFY_OK;
Herbert Xua9ea3fc2008-04-18 11:21:42 +08001308}
1309
Rick Jones8f9f4662011-10-19 08:10:59 +00001310static void virtnet_get_ringparam(struct net_device *dev,
1311 struct ethtool_ringparam *ring)
1312{
1313 struct virtnet_info *vi = netdev_priv(dev);
1314
Jason Wang986a4f42012-12-07 07:04:56 +00001315 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1316 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
Rick Jones8f9f4662011-10-19 08:10:59 +00001317 ring->rx_pending = ring->rx_max_pending;
1318 ring->tx_pending = ring->tx_max_pending;
Rick Jones8f9f4662011-10-19 08:10:59 +00001319}
1320
Rick Jones66846042011-11-14 14:17:08 +00001321
1322static void virtnet_get_drvinfo(struct net_device *dev,
1323 struct ethtool_drvinfo *info)
1324{
1325 struct virtnet_info *vi = netdev_priv(dev);
1326 struct virtio_device *vdev = vi->vdev;
1327
1328 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1329 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1330 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1331
1332}
1333
Jason Wangd73bcd22012-12-07 07:04:57 +00001334/* TODO: Eliminate OOO packets during switching */
1335static int virtnet_set_channels(struct net_device *dev,
1336 struct ethtool_channels *channels)
1337{
1338 struct virtnet_info *vi = netdev_priv(dev);
1339 u16 queue_pairs = channels->combined_count;
1340 int err;
1341
1342 /* We don't support separate rx/tx channels.
1343 * We don't allow setting 'other' channels.
1344 */
1345 if (channels->rx_count || channels->tx_count || channels->other_count)
1346 return -EINVAL;
1347
Amos Kongc18e9cd2014-04-18 13:45:41 +08001348 if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
Jason Wangd73bcd22012-12-07 07:04:57 +00001349 return -EINVAL;
1350
Wanlong Gao47be2472013-01-24 23:51:29 +00001351 get_online_cpus();
Jason Wangd73bcd22012-12-07 07:04:57 +00001352 err = virtnet_set_queues(vi, queue_pairs);
1353 if (!err) {
1354 netif_set_real_num_tx_queues(dev, queue_pairs);
1355 netif_set_real_num_rx_queues(dev, queue_pairs);
1356
Wanlong Gao8898c212013-01-24 23:51:30 +00001357 virtnet_set_affinity(vi);
Jason Wangd73bcd22012-12-07 07:04:57 +00001358 }
Wanlong Gao47be2472013-01-24 23:51:29 +00001359 put_online_cpus();
Jason Wangd73bcd22012-12-07 07:04:57 +00001360
1361 return err;
1362}
1363
1364static void virtnet_get_channels(struct net_device *dev,
1365 struct ethtool_channels *channels)
1366{
1367 struct virtnet_info *vi = netdev_priv(dev);
1368
1369 channels->combined_count = vi->curr_queue_pairs;
1370 channels->max_combined = vi->max_queue_pairs;
1371 channels->max_other = 0;
1372 channels->rx_count = 0;
1373 channels->tx_count = 0;
1374 channels->other_count = 0;
1375}
1376
Stephen Hemminger0fc0b732009-09-02 01:03:33 -07001377static const struct ethtool_ops virtnet_ethtool_ops = {
Rick Jones66846042011-11-14 14:17:08 +00001378 .get_drvinfo = virtnet_get_drvinfo,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001379 .get_link = ethtool_op_get_link,
Rick Jones8f9f4662011-10-19 08:10:59 +00001380 .get_ringparam = virtnet_get_ringparam,
Jason Wangd73bcd22012-12-07 07:04:57 +00001381 .set_channels = virtnet_set_channels,
1382 .get_channels = virtnet_get_channels,
Jacob Keller074c3582014-06-25 02:37:13 +00001383 .get_ts_info = ethtool_op_get_ts_info,
Herbert Xua9ea3fc2008-04-18 11:21:42 +08001384};
1385
Mark McLoughlin39da5812008-11-26 13:58:11 +00001386#define MIN_MTU 68
1387#define MAX_MTU 65535
1388
1389static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1390{
1391 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1392 return -EINVAL;
1393 dev->mtu = new_mtu;
1394 return 0;
1395}
1396
Stephen Hemminger76288b42009-01-06 10:44:22 -08001397static const struct net_device_ops virtnet_netdev = {
1398 .ndo_open = virtnet_open,
1399 .ndo_stop = virtnet_close,
1400 .ndo_start_xmit = start_xmit,
1401 .ndo_validate_addr = eth_validate_addr,
Alex Williamson9c46f6d2009-02-04 16:36:34 -08001402 .ndo_set_mac_address = virtnet_set_mac_address,
Alex Williamson2af76982009-02-04 09:02:40 +00001403 .ndo_set_rx_mode = virtnet_set_rx_mode,
Stephen Hemminger76288b42009-01-06 10:44:22 -08001404 .ndo_change_mtu = virtnet_change_mtu,
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001405 .ndo_get_stats64 = virtnet_stats,
Alex Williamson1824a982009-05-01 17:31:10 +00001406 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1407 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
Stephen Hemminger76288b42009-01-06 10:44:22 -08001408#ifdef CONFIG_NET_POLL_CONTROLLER
1409 .ndo_poll_controller = virtnet_netpoll,
1410#endif
Jason Wang91815632014-07-23 16:33:55 +08001411#ifdef CONFIG_NET_RX_BUSY_POLL
1412 .ndo_busy_poll = virtnet_busy_poll,
1413#endif
Stephen Hemminger76288b42009-01-06 10:44:22 -08001414};
1415
Jason Wang586d17c2012-04-11 20:43:52 +00001416static void virtnet_config_changed_work(struct work_struct *work)
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001417{
Jason Wang586d17c2012-04-11 20:43:52 +00001418 struct virtnet_info *vi =
1419 container_of(work, struct virtnet_info, config_work);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001420 u16 v;
1421
Rusty Russell855e0c52013-10-14 18:11:51 +10301422 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1423 struct virtio_net_config, status, &v) < 0)
Michael S. Tsirkin507613b2014-10-15 10:22:30 +10301424 return;
Jason Wang586d17c2012-04-11 20:43:52 +00001425
1426 if (v & VIRTIO_NET_S_ANNOUNCE) {
Amerigo Wangee89bab2012-08-09 22:14:56 +00001427 netdev_notify_peers(vi->dev);
Jason Wang586d17c2012-04-11 20:43:52 +00001428 virtnet_ack_link_announce(vi);
1429 }
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001430
1431 /* Ignore unknown (future) status bits */
1432 v &= VIRTIO_NET_S_LINK_UP;
1433
1434 if (vi->status == v)
Michael S. Tsirkin507613b2014-10-15 10:22:30 +10301435 return;
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001436
1437 vi->status = v;
1438
1439 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1440 netif_carrier_on(vi->dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001441 netif_tx_wake_all_queues(vi->dev);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001442 } else {
1443 netif_carrier_off(vi->dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001444 netif_tx_stop_all_queues(vi->dev);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001445 }
1446}
1447
1448static void virtnet_config_changed(struct virtio_device *vdev)
1449{
1450 struct virtnet_info *vi = vdev->priv;
1451
Tejun Heo3b07e9c2012-08-20 14:51:24 -07001452 schedule_work(&vi->config_work);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001453}
1454
Jason Wang986a4f42012-12-07 07:04:56 +00001455static void virtnet_free_queues(struct virtnet_info *vi)
1456{
Andrey Vagind4fb84e2013-12-05 18:36:21 +04001457 int i;
1458
Jason Wangab3971b2015-03-12 13:57:44 +08001459 for (i = 0; i < vi->max_queue_pairs; i++) {
1460 napi_hash_del(&vi->rq[i].napi);
Andrey Vagind4fb84e2013-12-05 18:36:21 +04001461 netif_napi_del(&vi->rq[i].napi);
Jason Wangab3971b2015-03-12 13:57:44 +08001462 }
Andrey Vagind4fb84e2013-12-05 18:36:21 +04001463
Jason Wang986a4f42012-12-07 07:04:56 +00001464 kfree(vi->rq);
1465 kfree(vi->sq);
1466}
1467
1468static void free_receive_bufs(struct virtnet_info *vi)
1469{
1470 int i;
1471
1472 for (i = 0; i < vi->max_queue_pairs; i++) {
1473 while (vi->rq[i].pages)
1474 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1475 }
1476}
1477
Michael Daltonfb518792014-01-16 22:23:26 -08001478static void free_receive_page_frags(struct virtnet_info *vi)
1479{
1480 int i;
1481 for (i = 0; i < vi->max_queue_pairs; i++)
1482 if (vi->rq[i].alloc_frag.page)
1483 put_page(vi->rq[i].alloc_frag.page);
1484}
1485
Jason Wang986a4f42012-12-07 07:04:56 +00001486static void free_unused_bufs(struct virtnet_info *vi)
1487{
1488 void *buf;
1489 int i;
1490
1491 for (i = 0; i < vi->max_queue_pairs; i++) {
1492 struct virtqueue *vq = vi->sq[i].vq;
1493 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1494 dev_kfree_skb(buf);
1495 }
1496
1497 for (i = 0; i < vi->max_queue_pairs; i++) {
1498 struct virtqueue *vq = vi->rq[i].vq;
1499
1500 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
Michael Daltonab7db912014-01-16 22:23:27 -08001501 if (vi->mergeable_rx_bufs) {
1502 unsigned long ctx = (unsigned long)buf;
1503 void *base = mergeable_ctx_to_buf_address(ctx);
1504 put_page(virt_to_head_page(base));
1505 } else if (vi->big_packets) {
Andrey Vaginfa9fac12013-12-05 18:36:20 +04001506 give_pages(&vi->rq[i], buf);
Michael Daltonab7db912014-01-16 22:23:27 -08001507 } else {
Jason Wang986a4f42012-12-07 07:04:56 +00001508 dev_kfree_skb(buf);
Michael Daltonab7db912014-01-16 22:23:27 -08001509 }
Jason Wang986a4f42012-12-07 07:04:56 +00001510 }
Jason Wang986a4f42012-12-07 07:04:56 +00001511 }
1512}
1513
Jason Wange9d74172012-12-07 07:04:55 +00001514static void virtnet_del_vqs(struct virtnet_info *vi)
1515{
1516 struct virtio_device *vdev = vi->vdev;
1517
Wanlong Gao8898c212013-01-24 23:51:30 +00001518 virtnet_clean_affinity(vi, -1);
Jason Wang986a4f42012-12-07 07:04:56 +00001519
Jason Wange9d74172012-12-07 07:04:55 +00001520 vdev->config->del_vqs(vdev);
Jason Wang986a4f42012-12-07 07:04:56 +00001521
1522 virtnet_free_queues(vi);
1523}
1524
1525static int virtnet_find_vqs(struct virtnet_info *vi)
1526{
1527 vq_callback_t **callbacks;
1528 struct virtqueue **vqs;
1529 int ret = -ENOMEM;
1530 int i, total_vqs;
1531 const char **names;
1532
1533 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1534 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1535 * possible control vq.
1536 */
1537 total_vqs = vi->max_queue_pairs * 2 +
1538 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1539
1540 /* Allocate space for find_vqs parameters */
1541 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1542 if (!vqs)
1543 goto err_vq;
1544 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1545 if (!callbacks)
1546 goto err_callback;
1547 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1548 if (!names)
1549 goto err_names;
1550
1551 /* Parameters for control virtqueue, if any */
1552 if (vi->has_cvq) {
1553 callbacks[total_vqs - 1] = NULL;
1554 names[total_vqs - 1] = "control";
1555 }
1556
1557 /* Allocate/initialize parameters for send/receive virtqueues */
1558 for (i = 0; i < vi->max_queue_pairs; i++) {
1559 callbacks[rxq2vq(i)] = skb_recv_done;
1560 callbacks[txq2vq(i)] = skb_xmit_done;
1561 sprintf(vi->rq[i].name, "input.%d", i);
1562 sprintf(vi->sq[i].name, "output.%d", i);
1563 names[rxq2vq(i)] = vi->rq[i].name;
1564 names[txq2vq(i)] = vi->sq[i].name;
1565 }
1566
1567 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1568 names);
1569 if (ret)
1570 goto err_find;
1571
1572 if (vi->has_cvq) {
1573 vi->cvq = vqs[total_vqs - 1];
1574 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
Patrick McHardyf6469682013-04-19 02:04:27 +00001575 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
Jason Wang986a4f42012-12-07 07:04:56 +00001576 }
1577
1578 for (i = 0; i < vi->max_queue_pairs; i++) {
1579 vi->rq[i].vq = vqs[rxq2vq(i)];
1580 vi->sq[i].vq = vqs[txq2vq(i)];
1581 }
1582
1583 kfree(names);
1584 kfree(callbacks);
1585 kfree(vqs);
1586
1587 return 0;
1588
1589err_find:
1590 kfree(names);
1591err_names:
1592 kfree(callbacks);
1593err_callback:
1594 kfree(vqs);
1595err_vq:
1596 return ret;
1597}
1598
1599static int virtnet_alloc_queues(struct virtnet_info *vi)
1600{
1601 int i;
1602
1603 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1604 if (!vi->sq)
1605 goto err_sq;
1606 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
Amerigo Wang008d4272012-12-10 02:24:08 +00001607 if (!vi->rq)
Jason Wang986a4f42012-12-07 07:04:56 +00001608 goto err_rq;
1609
1610 INIT_DELAYED_WORK(&vi->refill, refill_work);
1611 for (i = 0; i < vi->max_queue_pairs; i++) {
1612 vi->rq[i].pages = NULL;
1613 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1614 napi_weight);
Jason Wang91815632014-07-23 16:33:55 +08001615 napi_hash_add(&vi->rq[i].napi);
Jason Wang986a4f42012-12-07 07:04:56 +00001616
1617 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
Johannes Berg5377d7582015-08-19 09:48:40 +02001618 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
Jason Wang986a4f42012-12-07 07:04:56 +00001619 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1620 }
1621
1622 return 0;
1623
1624err_rq:
1625 kfree(vi->sq);
1626err_sq:
1627 return -ENOMEM;
Jason Wange9d74172012-12-07 07:04:55 +00001628}
1629
Amit Shah3f9c10b2011-12-22 16:58:31 +05301630static int init_vqs(struct virtnet_info *vi)
1631{
Jason Wang986a4f42012-12-07 07:04:56 +00001632 int ret;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301633
Jason Wang986a4f42012-12-07 07:04:56 +00001634 /* Allocate send & receive queues */
1635 ret = virtnet_alloc_queues(vi);
1636 if (ret)
1637 goto err;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301638
Jason Wang986a4f42012-12-07 07:04:56 +00001639 ret = virtnet_find_vqs(vi);
1640 if (ret)
1641 goto err_free;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301642
Wanlong Gao47be2472013-01-24 23:51:29 +00001643 get_online_cpus();
Wanlong Gao8898c212013-01-24 23:51:30 +00001644 virtnet_set_affinity(vi);
Wanlong Gao47be2472013-01-24 23:51:29 +00001645 put_online_cpus();
1646
Amit Shah3f9c10b2011-12-22 16:58:31 +05301647 return 0;
Jason Wang986a4f42012-12-07 07:04:56 +00001648
1649err_free:
1650 virtnet_free_queues(vi);
1651err:
1652 return ret;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301653}
1654
Michael Daltonfbf28d72014-01-16 22:23:30 -08001655#ifdef CONFIG_SYSFS
1656static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
1657 struct rx_queue_attribute *attribute, char *buf)
1658{
1659 struct virtnet_info *vi = netdev_priv(queue->dev);
1660 unsigned int queue_index = get_netdev_rx_queue_index(queue);
Johannes Berg5377d7582015-08-19 09:48:40 +02001661 struct ewma_pkt_len *avg;
Michael Daltonfbf28d72014-01-16 22:23:30 -08001662
1663 BUG_ON(queue_index >= vi->max_queue_pairs);
1664 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
1665 return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
1666}
1667
1668static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
1669 __ATTR_RO(mergeable_rx_buffer_size);
1670
1671static struct attribute *virtio_net_mrg_rx_attrs[] = {
1672 &mergeable_rx_buffer_size_attribute.attr,
1673 NULL
1674};
1675
1676static const struct attribute_group virtio_net_mrg_rx_group = {
1677 .name = "virtio_net",
1678 .attrs = virtio_net_mrg_rx_attrs
1679};
1680#endif
1681
Jason Wang892d6eb2014-11-20 17:03:05 +08001682static bool virtnet_fail_on_feature(struct virtio_device *vdev,
1683 unsigned int fbit,
1684 const char *fname, const char *dname)
1685{
1686 if (!virtio_has_feature(vdev, fbit))
1687 return false;
1688
1689 dev_err(&vdev->dev, "device advertises feature %s but not %s",
1690 fname, dname);
1691
1692 return true;
1693}
1694
1695#define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
1696 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
1697
1698static bool virtnet_validate_features(struct virtio_device *vdev)
1699{
1700 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
1701 (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
1702 "VIRTIO_NET_F_CTRL_VQ") ||
1703 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
1704 "VIRTIO_NET_F_CTRL_VQ") ||
1705 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
1706 "VIRTIO_NET_F_CTRL_VQ") ||
1707 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
1708 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
1709 "VIRTIO_NET_F_CTRL_VQ"))) {
1710 return false;
1711 }
1712
1713 return true;
1714}
1715
Rusty Russell296f96f2007-10-22 11:03:37 +10001716static int virtnet_probe(struct virtio_device *vdev)
1717{
Jason Wang986a4f42012-12-07 07:04:56 +00001718 int i, err;
Rusty Russell296f96f2007-10-22 11:03:37 +10001719 struct net_device *dev;
1720 struct virtnet_info *vi;
Jason Wang986a4f42012-12-07 07:04:56 +00001721 u16 max_queue_pairs;
1722
Michael S. Tsirkin6ba42242015-01-12 16:23:37 +02001723 if (!vdev->config->get) {
1724 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1725 __func__);
1726 return -EINVAL;
1727 }
1728
Jason Wang892d6eb2014-11-20 17:03:05 +08001729 if (!virtnet_validate_features(vdev))
1730 return -EINVAL;
1731
Jason Wang986a4f42012-12-07 07:04:56 +00001732 /* Find if host supports multiqueue virtio_net device */
Rusty Russell855e0c52013-10-14 18:11:51 +10301733 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1734 struct virtio_net_config,
1735 max_virtqueue_pairs, &max_queue_pairs);
Jason Wang986a4f42012-12-07 07:04:56 +00001736
1737 /* We need at least 2 queue's */
1738 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1739 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1740 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1741 max_queue_pairs = 1;
Rusty Russell296f96f2007-10-22 11:03:37 +10001742
1743 /* Allocate ourselves a network device with room for our info */
Jason Wang986a4f42012-12-07 07:04:56 +00001744 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
Rusty Russell296f96f2007-10-22 11:03:37 +10001745 if (!dev)
1746 return -ENOMEM;
1747
1748 /* Set up network device as normal. */
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +00001749 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
Stephen Hemminger76288b42009-01-06 10:44:22 -08001750 dev->netdev_ops = &virtnet_netdev;
Rusty Russell296f96f2007-10-22 11:03:37 +10001751 dev->features = NETIF_F_HIGHDMA;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001752
Wilfried Klaebe7ad24ea2014-05-11 00:12:32 +00001753 dev->ethtool_ops = &virtnet_ethtool_ops;
Rusty Russell296f96f2007-10-22 11:03:37 +10001754 SET_NETDEV_DEV(dev, &vdev->dev);
1755
1756 /* Do we support "hardware" checksums? */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001757 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
Rusty Russell296f96f2007-10-22 11:03:37 +10001758 /* This opens up the world of extra features. */
Jason Wang48900cb2015-08-05 10:34:04 +08001759 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
Michał Mirosław98e778c2011-03-31 01:01:35 +00001760 if (csum)
Jason Wang48900cb2015-08-05 10:34:04 +08001761 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
Michał Mirosław98e778c2011-03-31 01:01:35 +00001762
1763 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001764 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
Rusty Russell34a48572008-02-04 23:50:02 -05001765 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1766 }
Rusty Russell5539ae92008-05-02 21:50:46 -05001767 /* Individual feature bits: what can host handle? */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001768 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1769 dev->hw_features |= NETIF_F_TSO;
1770 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1771 dev->hw_features |= NETIF_F_TSO6;
1772 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1773 dev->hw_features |= NETIF_F_TSO_ECN;
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001774 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1775 dev->hw_features |= NETIF_F_UFO;
Michał Mirosław98e778c2011-03-31 01:01:35 +00001776
Jason Wang41f2f122014-12-24 11:03:52 +08001777 dev->features |= NETIF_F_GSO_ROBUST;
1778
Michał Mirosław98e778c2011-03-31 01:01:35 +00001779 if (gso)
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001780 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
Michał Mirosław98e778c2011-03-31 01:01:35 +00001781 /* (!csum && gso) case will be fixed by register_netdev() */
Rusty Russell296f96f2007-10-22 11:03:37 +10001782 }
Thomas Huth4f491292013-08-27 17:09:02 +02001783 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1784 dev->features |= NETIF_F_RXCSUM;
Rusty Russell296f96f2007-10-22 11:03:37 +10001785
Jason Wang4fda8302013-04-10 23:32:21 +00001786 dev->vlan_features = dev->features;
1787
Rusty Russell296f96f2007-10-22 11:03:37 +10001788 /* Configuration may specify what MAC to use. Otherwise random. */
Rusty Russell855e0c52013-10-14 18:11:51 +10301789 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1790 virtio_cread_bytes(vdev,
1791 offsetof(struct virtio_net_config, mac),
1792 dev->dev_addr, dev->addr_len);
1793 else
Danny Kukawkaf2cedb62012-02-15 06:45:39 +00001794 eth_hw_addr_random(dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001795
1796 /* Set up our device-specific information */
1797 vi = netdev_priv(dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001798 vi->dev = dev;
1799 vi->vdev = vdev;
Christian Borntraegerd9d5dcc2008-02-18 10:02:51 +01001800 vdev->priv = vi;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001801 vi->stats = alloc_percpu(struct virtnet_stats);
1802 err = -ENOMEM;
1803 if (vi->stats == NULL)
1804 goto free;
1805
John Stultz827da442013-10-07 15:51:58 -07001806 for_each_possible_cpu(i) {
1807 struct virtnet_stats *virtnet_stats;
1808 virtnet_stats = per_cpu_ptr(vi->stats, i);
1809 u64_stats_init(&virtnet_stats->tx_syncp);
1810 u64_stats_init(&virtnet_stats->rx_syncp);
1811 }
1812
Jason Wang586d17c2012-04-11 20:43:52 +00001813 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
Rusty Russell296f96f2007-10-22 11:03:37 +10001814
Herbert Xu97402b92008-04-18 11:24:27 +08001815 /* If we can receive ANY GSO packets, we must allocate large ones. */
Joe Perches8e95a202009-12-03 07:58:21 +00001816 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1817 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001818 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
1819 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
Herbert Xu97402b92008-04-18 11:24:27 +08001820 vi->big_packets = true;
1821
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -08001822 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1823 vi->mergeable_rx_bufs = true;
1824
Michael S. Tsirkind04302b2014-10-24 00:24:03 +03001825 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
1826 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
Michael S. Tsirkin012873d2014-10-24 16:55:57 +03001827 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1828 else
1829 vi->hdr_len = sizeof(struct virtio_net_hdr);
1830
Michael S. Tsirkin75993302015-07-15 15:26:19 +03001831 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
1832 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
Michael S. Tsirkine7428e92013-07-25 10:20:23 +09301833 vi->any_header_sg = true;
1834
Jason Wang986a4f42012-12-07 07:04:56 +00001835 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1836 vi->has_cvq = true;
1837
Michael S. Tsirkin012873d2014-10-24 16:55:57 +03001838 if (vi->any_header_sg)
1839 dev->needed_headroom = vi->hdr_len;
Zhangjie \(HZ\)6ebbc1a2014-04-29 18:43:22 +08001840
Jason Wang986a4f42012-12-07 07:04:56 +00001841 /* Use single tx/rx queue pair as default */
1842 vi->curr_queue_pairs = 1;
1843 vi->max_queue_pairs = max_queue_pairs;
1844
1845 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
Amit Shah3f9c10b2011-12-22 16:58:31 +05301846 err = init_vqs(vi);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001847 if (err)
Jason Wang9bb8ca82013-11-05 18:19:45 +08001848 goto free_stats;
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001849
Michael Daltonfbf28d72014-01-16 22:23:30 -08001850#ifdef CONFIG_SYSFS
1851 if (vi->mergeable_rx_bufs)
1852 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
1853#endif
Zhi Yong Wu0f13b662013-11-18 21:19:27 +08001854 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1855 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
Jason Wang986a4f42012-12-07 07:04:56 +00001856
Rusty Russell296f96f2007-10-22 11:03:37 +10001857 err = register_netdev(dev);
1858 if (err) {
1859 pr_debug("virtio_net: registering device failed\n");
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001860 goto free_vqs;
Rusty Russell296f96f2007-10-22 11:03:37 +10001861 }
Rusty Russellb3369c12008-02-04 23:50:02 -05001862
Michael S. Tsirkin4baf1e32014-10-15 10:22:30 +10301863 virtio_device_ready(vdev);
1864
Rusty Russellb3369c12008-02-04 23:50:02 -05001865 /* Last of all, set up some receive buffers. */
Sasha Levin55257d72013-04-29 12:00:08 +09301866 for (i = 0; i < vi->curr_queue_pairs; i++) {
Michael S. Tsirkin946fa562014-10-24 00:12:10 +03001867 try_fill_recv(vi, &vi->rq[i], GFP_KERNEL);
Rusty Russellb3369c12008-02-04 23:50:02 -05001868
Jason Wang986a4f42012-12-07 07:04:56 +00001869 /* If we didn't even get one input buffer, we're useless. */
Jason Wangbe121f42014-01-16 14:45:24 +08001870 if (vi->rq[i].vq->num_free ==
1871 virtqueue_get_vring_size(vi->rq[i].vq)) {
Jason Wang986a4f42012-12-07 07:04:56 +00001872 free_unused_bufs(vi);
1873 err = -ENOMEM;
1874 goto free_recv_bufs;
1875 }
Rusty Russellb3369c12008-02-04 23:50:02 -05001876 }
1877
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001878 vi->nb.notifier_call = &virtnet_cpu_callback;
1879 err = register_hotcpu_notifier(&vi->nb);
1880 if (err) {
1881 pr_debug("virtio_net: registering cpu notifier failed\n");
1882 goto free_recv_bufs;
1883 }
1884
Jason Wang167c25e2010-11-10 14:45:41 +00001885 /* Assume link up if device can't report link status,
1886 otherwise get link status from config. */
1887 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1888 netif_carrier_off(dev);
Tejun Heo3b07e9c2012-08-20 14:51:24 -07001889 schedule_work(&vi->config_work);
Jason Wang167c25e2010-11-10 14:45:41 +00001890 } else {
1891 vi->status = VIRTIO_NET_S_LINK_UP;
1892 netif_carrier_on(dev);
1893 }
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001894
Jason Wang986a4f42012-12-07 07:04:56 +00001895 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1896 dev->name, max_queue_pairs);
1897
Rusty Russell296f96f2007-10-22 11:03:37 +10001898 return 0;
1899
Jason Wang986a4f42012-12-07 07:04:56 +00001900free_recv_bufs:
Michael S. Tsirkin02465552014-10-15 10:22:31 +10301901 vi->vdev->config->reset(vdev);
1902
Jason Wang986a4f42012-12-07 07:04:56 +00001903 free_receive_bufs(vi);
Rusty Russellb3369c12008-02-04 23:50:02 -05001904 unregister_netdev(dev);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001905free_vqs:
Jason Wang986a4f42012-12-07 07:04:56 +00001906 cancel_delayed_work_sync(&vi->refill);
Michael Daltonfb518792014-01-16 22:23:26 -08001907 free_receive_page_frags(vi);
Jason Wange9d74172012-12-07 07:04:55 +00001908 virtnet_del_vqs(vi);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001909free_stats:
1910 free_percpu(vi->stats);
Rusty Russell296f96f2007-10-22 11:03:37 +10001911free:
1912 free_netdev(dev);
1913 return err;
1914}
1915
Amit Shah04486ed2011-12-22 16:58:32 +05301916static void remove_vq_common(struct virtnet_info *vi)
Rusty Russell296f96f2007-10-22 11:03:37 +10001917{
Amit Shah04486ed2011-12-22 16:58:32 +05301918 vi->vdev->config->reset(vi->vdev);
Shirley Ma830a8a92010-02-08 14:14:42 +00001919
1920 /* Free unused buffers in both send and recv, if any. */
Shirley Ma9ab86bb2010-01-29 03:20:04 +00001921 free_unused_bufs(vi);
Rusty Russellfb6813f2008-07-25 12:06:01 -05001922
Jason Wang986a4f42012-12-07 07:04:56 +00001923 free_receive_bufs(vi);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001924
Michael Daltonfb518792014-01-16 22:23:26 -08001925 free_receive_page_frags(vi);
1926
Jason Wang986a4f42012-12-07 07:04:56 +00001927 virtnet_del_vqs(vi);
Amit Shah04486ed2011-12-22 16:58:32 +05301928}
1929
Bill Pemberton8cc085d2012-12-03 09:24:15 -05001930static void virtnet_remove(struct virtio_device *vdev)
Amit Shah04486ed2011-12-22 16:58:32 +05301931{
1932 struct virtnet_info *vi = vdev->priv;
1933
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001934 unregister_hotcpu_notifier(&vi->nb);
1935
Michael S. Tsirkin102a2782014-10-15 10:22:29 +10301936 /* Make sure no work handler is accessing the device. */
1937 flush_work(&vi->config_work);
Jason Wang586d17c2012-04-11 20:43:52 +00001938
Amit Shah04486ed2011-12-22 16:58:32 +05301939 unregister_netdev(vi->dev);
1940
1941 remove_vq_common(vi);
Rusty Russellfb6813f2008-07-25 12:06:01 -05001942
Krishna Kumar2e66f552011-07-20 03:56:02 +00001943 free_percpu(vi->stats);
Rusty Russell74b25532007-11-19 11:20:42 -05001944 free_netdev(vi->dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001945}
1946
Aaron Lu89107002013-09-17 09:25:23 +09301947#ifdef CONFIG_PM_SLEEP
Amit Shah0741bcb2011-12-22 16:58:33 +05301948static int virtnet_freeze(struct virtio_device *vdev)
1949{
1950 struct virtnet_info *vi = vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +00001951 int i;
Amit Shah0741bcb2011-12-22 16:58:33 +05301952
Jason Wangec9debb2013-10-29 15:11:07 +08001953 unregister_hotcpu_notifier(&vi->nb);
1954
Michael S. Tsirkin102a2782014-10-15 10:22:29 +10301955 /* Make sure no work handler is accessing the device */
1956 flush_work(&vi->config_work);
Jason Wang586d17c2012-04-11 20:43:52 +00001957
Amit Shah0741bcb2011-12-22 16:58:33 +05301958 netif_device_detach(vi->dev);
1959 cancel_delayed_work_sync(&vi->refill);
1960
Jason Wang91815632014-07-23 16:33:55 +08001961 if (netif_running(vi->dev)) {
Jason Wangab3971b2015-03-12 13:57:44 +08001962 for (i = 0; i < vi->max_queue_pairs; i++)
Jason Wang986a4f42012-12-07 07:04:56 +00001963 napi_disable(&vi->rq[i].napi);
Jason Wang91815632014-07-23 16:33:55 +08001964 }
Amit Shah0741bcb2011-12-22 16:58:33 +05301965
1966 remove_vq_common(vi);
1967
1968 return 0;
1969}
1970
1971static int virtnet_restore(struct virtio_device *vdev)
1972{
1973 struct virtnet_info *vi = vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +00001974 int err, i;
Amit Shah0741bcb2011-12-22 16:58:33 +05301975
1976 err = init_vqs(vi);
1977 if (err)
1978 return err;
1979
Michael S. Tsirkine53fbd12014-10-15 10:22:32 +10301980 virtio_device_ready(vdev);
1981
Jason Wang6cd4ce02013-12-30 11:34:40 +08001982 if (netif_running(vi->dev)) {
1983 for (i = 0; i < vi->curr_queue_pairs; i++)
Michael S. Tsirkin946fa562014-10-24 00:12:10 +03001984 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
Jason Wang6cd4ce02013-12-30 11:34:40 +08001985 schedule_delayed_work(&vi->refill, 0);
1986
Jason Wang986a4f42012-12-07 07:04:56 +00001987 for (i = 0; i < vi->max_queue_pairs; i++)
1988 virtnet_napi_enable(&vi->rq[i]);
Jason Wang6cd4ce02013-12-30 11:34:40 +08001989 }
Amit Shah0741bcb2011-12-22 16:58:33 +05301990
1991 netif_device_attach(vi->dev);
1992
Jason Wang35ed1592013-10-15 11:18:59 +08001993 rtnl_lock();
Jason Wang986a4f42012-12-07 07:04:56 +00001994 virtnet_set_queues(vi, vi->curr_queue_pairs);
Jason Wang35ed1592013-10-15 11:18:59 +08001995 rtnl_unlock();
Jason Wang986a4f42012-12-07 07:04:56 +00001996
Jason Wangec9debb2013-10-29 15:11:07 +08001997 err = register_hotcpu_notifier(&vi->nb);
1998 if (err)
1999 return err;
2000
Amit Shah0741bcb2011-12-22 16:58:33 +05302001 return 0;
2002}
2003#endif
2004
Rusty Russell296f96f2007-10-22 11:03:37 +10002005static struct virtio_device_id id_table[] = {
2006 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
2007 { 0 },
2008};
2009
Rusty Russellc45a6812008-05-02 21:50:50 -05002010static unsigned int features[] = {
Mark McLoughlin5e4fe5c2008-07-08 17:10:42 +10002011 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
2012 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05002013 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
Herbert Xu97402b92008-04-18 11:24:27 +08002014 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05002015 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
Alex Williamson2a41f712009-02-04 09:02:34 +00002016 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
Alex Williamson0bde95692009-02-04 09:02:50 +00002017 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
Jason Wang986a4f42012-12-07 07:04:56 +00002018 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
Amos Kong7e58d5a2013-01-21 01:17:23 +00002019 VIRTIO_NET_F_CTRL_MAC_ADDR,
Michael S. Tsirkine7428e92013-07-25 10:20:23 +09302020 VIRTIO_F_ANY_LAYOUT,
Rusty Russellc45a6812008-05-02 21:50:50 -05002021};
2022
Uwe Kleine-König22402522009-11-05 01:32:44 -08002023static struct virtio_driver virtio_net_driver = {
Rusty Russellc45a6812008-05-02 21:50:50 -05002024 .feature_table = features,
2025 .feature_table_size = ARRAY_SIZE(features),
Rusty Russell296f96f2007-10-22 11:03:37 +10002026 .driver.name = KBUILD_MODNAME,
2027 .driver.owner = THIS_MODULE,
2028 .id_table = id_table,
2029 .probe = virtnet_probe,
Bill Pemberton8cc085d2012-12-03 09:24:15 -05002030 .remove = virtnet_remove,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08002031 .config_changed = virtnet_config_changed,
Aaron Lu89107002013-09-17 09:25:23 +09302032#ifdef CONFIG_PM_SLEEP
Amit Shah0741bcb2011-12-22 16:58:33 +05302033 .freeze = virtnet_freeze,
2034 .restore = virtnet_restore,
2035#endif
Rusty Russell296f96f2007-10-22 11:03:37 +10002036};
2037
Rusty Russellb2a17022013-02-13 16:59:28 +10302038module_virtio_driver(virtio_net_driver);
Rusty Russell296f96f2007-10-22 11:03:37 +10002039
2040MODULE_DEVICE_TABLE(virtio, id_table);
2041MODULE_DESCRIPTION("Virtio network driver");
2042MODULE_LICENSE("GPL");